Prognosis Prediction Model For Septic Patients Based On Nuclear Magnetic Resonance Metabolomics

Objective:Sepsis is one of the leading causes of death in patients who admitted to intensive care units.Due to the complex pathophysiological mechanism and the rapid deterioration of the disease,the prognosis of patients with sepsis is difficult to predict.The aim of the study is to establish a prognostic model for septic patients based on proton nuclear magnetic resonance（~1H NMR）metabolomics.Firstly,the author adopted an evidence-based approach to assess the methodological quality of existing studies.The predictive performance and biomarkers of these studies were systematically evaluated,thus to explore the shortcomings and challenges of the current study.Based on the discoveries by the above works,the author designed and conducted a prospective cohort combined with NMR spectroscopy metabolomics.Consequently,the study would like to discover new biomarkers,combined with time-series analysis.This model enables clinical practitioners to monitor the evolution of sepsis dynamically.Methods:The study began with a systematic review of existing studies on sepsis metabolic marker to comprehensively screen for potential metabolic markers that have been reported to predict patient outcomes.Then a prospective cohort of septic patients was established.Patient’s serum samples were collected at multiple time points for acquisition of 1H metabolic profile information using an 800 MHz NMR.According to the 28-day survival of patients,patients were divided into two cohorts:survival and death.Finally,a prognosis prediction model for septic patients was established through these differential metabolites.Results:Through literature search,1261 articles were initially screened,and 14studies met inclusion criteria and included for systematic review.The methodological evaluation indicated that 10 out of 14 studies were of low quality,indicating that the quality of research in this field was generally not high.Previous studies mentioned 48potential prognostic metabolites,the main chemical classes being amino acids and their derivatives（24%）,lipids and lipid-like molecules（39%）,and organic acids and their derivatives（30%）.Metabolic pathway disorders were mainly occurred in pathways such as amino acid,phospholipid and tricarboxylic acid cycles.The prospective cohort study included 51 sepsis patients,of whom 23 died and 28 survived.In the study,162 samples were tested,and 23 differential metabolites were initially screened.After partial least squares discriminant analysis,the results showed that the biomarkers set composed of acetoacetate,phosphocreatine,4-hydroxyphenylacetic acid,tyrosine and 2’-deoxyuridine could accurately predict the prognosis of patients with sepsis.The predictive power was good,with an Area Under the Receiver Operating Characteristic Curve（AUROC）of 0.85,95%CI:（0.723-0.988）.Conclusion:The results of the study suggest that metabolic biomarkers are potential prognostic indicators for sepsis,but the overall predictive power of any single marker is not high.Prediction performance can be significantly improved by combining multiple metabolic markers into biomarkers set.